Organization Connectors
Organization connectors (also called org-level connectors) are set up once by an administrator or developer and shared across the organization. They read data from your source systems into Abacus.AI, where it becomes a dataset or feature group that you can train models on, build GenAI document sets from, and attach to Custom Chatbots.
If you instead want each person to query their own account live from chat, you need a user connector. The Connectors overview compares the two.
How organization connectors work​
- An admin or developer creates the connector in the Connected Services Dashboard (click your profile picture in the top right, then Manage Connectors). The connector stores shared credentials: a database user, an IAM role, an OAuth login by the admin, or a service principal or certificate.
- Abacus.AI verifies the connection so you can confirm it reaches the source before anyone depends on it.
- Developers read from the connector inside a project, by creating a dataset with Read from External Service and choosing the connector. The dataset can be refreshed on a schedule, or kept continuously in sync with Realtime Synced Datasets.
- The data is used to train models or to build the vector store behind a Custom Chatbot. Everyone who can use the resulting model sees the same data, unless you use a Permission Aware Connector.
Organization connectors are by default available only to developers who have access to the development environment of the platform. You can restrict them further with the RBAC Module.
When to use an organization connector​
| Use an organization connector when… | Use a user connector instead when… |
|---|---|
| You are training an ML model on a table or set of files | Each user should query the source under their own permissions |
| You are building a RAG chatbot over a defined document set | You want a live answer in chat without building a project |
| The data must be refreshed on a schedule or synced continuously | Freshness on every single question matters more than a curated copy |
| A developer needs to control what is ingested and how it is transformed | An end user wants to connect their own account in a couple of clicks |
Connector types​
| Type | What it reads | Examples |
|---|---|---|
| Database Connectors | Tables and SQL query results from relational and analytical databases | Snowflake, BigQuery, Athena, Oracle, MySQL, JDBC/ODBC |
| File Connectors | Files and folders from object storage and file transfer services | AWS S3, Google Cloud Storage, Azure Storage, SFTP |
| Application Connectors | Documents, records and tickets from business applications | SharePoint, OneDrive, Google Drive, Confluence, Jira, Zendesk, HubSpot |
| Messaging Connectors | Chat platforms, both as a data source and as a place to deploy chatbots | Slack, Microsoft Teams, Teams Transcripts |
| Permission Aware Connectors | The same sources as above, while preserving each document's access rights | SharePoint, Confluence, Jira, Teams Transcripts |
| Realtime Synced Datasets | Continuous sync from a supported connector instead of scheduled refreshes | Jira, Confluence, SharePoint, Teams |
Config connectors​
Some user connectors depend on a small organization-level record called a config connector (the Config Only toggle on the connector form). A config connector does not ingest any data. It stores the OAuth application, credentials or scopes that every user-level connection for that service inherits, for example the Tableau Connected App, the NetSuite Integration Record, or the scopes an Outlook or Smartsheet connection should request.
Config connectors are created by an admin in the Connected Services Dashboard, but they belong to the user connector flow. See Scopes and config connectors in the User Connectors guide.
Before you start​
- Microsoft connectors need tenant consent. For OneDrive, SharePoint, Outlook, Microsoft Teams, Azure Storage and Azure SQL, a Microsoft Entra admin must create the connector first and tick "Consent on behalf of your organization" on Microsoft's consent screen. See Microsoft Entra Tenant Consent.
- Sources behind a firewall need the Abacus.AI egress IPs allowlisted.
- Custom OAuth flows must register
https://abacus.ai/oauth/callbackas the redirect URI on your own OAuth application.